Ask, then check the work.
A natural-language answer you cannot verify is a rumour. Vemon AI returns the query it ran next to the result, so you can read it, edit it, and run it yourself.
Intended shape
How it will look
The schema and shapes here match what the platform already serves. Only the endpoint is missing.
curl "https://api.vemon.io/v1/ai/ask" \
-H "Authorization: Bearer vm_live_…" \
-d '{"q":"which procedure has the widest charge spread?"}'How it works
Design decisions, and why
Where a choice costs convenience, the reason is stated rather than hidden.
The query is part of the output
Every response includes the SQL it executed. Copy it, change it, run it directly. An answer you cannot reproduce is not an answer.
Grounded in query results
Answers come from executing a query against the loaded datasets, not from a model's recollection of what the data might contain.
Refuses rather than guesses
If a question cannot be answered from what is loaded, it says so and names what is missing.
Auditable
Every AI query is logged like any other request, with the generated SQL retained alongside it.
Use cases
What people build with it
Internal analytics assistants
Let a non-technical colleague ask a question and hand an analyst the query behind it.
Agent tooling
Give an agent a data source whose answers carry their own provenance and can be checked.
Exploration
Find the shape of a dataset before writing the query you actually need.
Specification
The details
| Output | Answer, generated SQL, and the rows it ran on. |
|---|---|
| Grounding | Executed queries only — no unverified generation. |
| Refusal | Explicit when the question exceeds loaded data. |
| Logging | Query and result retained for audit. |
| Status | Not built. |
Questions
Asked most often
- Is Vemon AI available today?
- No. This page describes the intended design so it can be argued with before it is written. The endpoint returns 404.
- Will answers be verifiable?
- That is the entire premise. Every response returns the query it executed, so you can check the reasoning rather than trusting a summary.
- What happens if the data cannot answer the question?
- It refuses and names the gap, rather than producing a plausible number from nothing.
Also on the platform
The other surfaces
The API is live today.
While this is being built, the REST API already serves the same data.